Re: [PATCH] nfc: virtual_ncidev: Add variable to check if ndev is running

[Krzysztof Kozlowski]

On 20/11/2023 05:47, Bongsu Jeon wrote:
> 
> On 20/11/2023 01:47, Nguyen Dinh Phi wrote:
> 
>> syzbot reported an memory leak that happens when an skb is add to
>> send_buff after virtual nci closed.
>> This patch adds a variable to track if the ndev is running before
>> handling new skb in send function.
>>
>> Reported-by: syzbot+6eb09d75211863f15e3e@xxxxxxxxxxxxxxxxxxxxxxxxx
>> Closes: https://lore.kernel.org/lkml/00000000000075472b06007df4fb@xxxxxxxxxx
>> Signed-off-by: Nguyen Dinh Phi <phind.uet@xxxxxxxxx>
>> ---
>>  drivers/nfc/virtual_ncidev.c | 9 +++++++--
>>  1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/nfc/virtual_ncidev.c b/drivers/nfc/virtual_ncidev.c
>> index b027be0b0b6f..ac8226db54e2 100644
>> --- a/drivers/nfc/virtual_ncidev.c
>> +++ b/drivers/nfc/virtual_ncidev.c
>> @@ -20,26 +20,31 @@
>>                                   NFC_PROTO_ISO14443_MASK | \
>>                                   NFC_PROTO_ISO14443_B_MASK | \
>>                                   NFC_PROTO_ISO15693_MASK)
>> +#define NCIDEV_RUNNING 0
> This define isn't used.
> 
>>  
>>  struct virtual_nci_dev {
>>          struct nci_dev *ndev;
>>          struct mutex mtx;
>>          struct sk_buff *send_buff;
>>          struct wait_queue_head wq;
>> +        bool running;
>>  };
>>  
>>  static int virtual_nci_open(struct nci_dev *ndev)
>>  {
>> +        struct virtual_nci_dev *vdev = nci_get_drvdata(ndev);
>> +
>> +        vdev->running = true;
>>          return 0;
>>  }
>>  
>>  static int virtual_nci_close(struct nci_dev *ndev)
>>  {
>>          struct virtual_nci_dev *vdev = nci_get_drvdata(ndev);
>> -
>>          mutex_lock(&vdev->mtx);
>>          kfree_skb(vdev->send_buff);
>>          vdev->send_buff = NULL;
>> +        vdev->running = false;
>>          mutex_unlock(&vdev->mtx);
>>  
>>          return 0;
>> @@ -50,7 +55,7 @@ static int virtual_nci_send(struct nci_dev *ndev, struct sk_buff *skb)
>>          struct virtual_nci_dev *vdev = nci_get_drvdata(ndev);
>>  
>>          mutex_lock(&vdev->mtx);
>> -        if (vdev->send_buff) {
>> +        if (vdev->send_buff || !vdev->running) {
> 
> Dear Krzysztof,
> 
> I agree this defensive code.
> But i think NFC submodule has to avoid this situation.(calling send function of closed nci_dev)
> Could you check this?

This code looks not effective. At this point vdev->send_buff is always
false, so the additional check would not bring any value.

I don't see this fixing anything. Syzbot also does not seem to agree.

Nguyen, please test your patches against syzbot *before* sending them.
If you claim this fixes the report, please provide me the link to syzbot
test results confirming it is fixed.

I looked at syzbot dashboard and do not see this issue fixed with this
patch.

Best regards,
Krzysztof